- November 05, 2026
- By John Tucker
James Martin raced to the woman splayed on the ground, one arm severed above the wrist. He grabbed scissors from his medic’s bag and sheared off her clothes, preparing to apply a tourniquet, when a geyser of blood erupted from her upper thigh.
“Don’t let me die,” whimpered the woman, her chest heaving.
“I’m trying not to, lady,” Martin said, plugging her wound with a gloved finger and explaining she’d been in a car crash.
Martin scanned the scene: a 13-vehicle pileup in which a city bus rear-ended a school bus and a garbage truck flattened a sedan. Two dozen victims were hurt, some semiconscious and wailing, others staggering around like zombies. “I need some help! He’s stuck in the truck and can’t breathe!” a man cried in the smoky distance.
In reality, no tragedy had occurred. The smashed-up vehicles were part of a simulation at the Fairfax Urban Search and Rescue’s training facility in Lorton, Va., and the victims were role players plastered with horror-film-grade blood, along with two manikins — including Tammy, the woman with the groin injury — who can talk via AI and appear to breathe, pulsate and bleed.
A crowd of EMTs, police officers and U.S. military representatives had assembled on this mild summer morning to watch a live experiment run by a University of Maryland robotics team to showcase an artificial intelligence system it developed to aid the lifesaving efforts of first responders like Martin. With limited numbers, medics often can’t locate, let alone treat, everyone at once at a mass casualty scene.
In the 10 minutes before Martin and his partner, Vienna, entered the simulation, two drones and three autonomous four-legged robot “dogs” had pinpointed and photographed victims, used an algorithm to assess their injuries, then fed the results to the medics, via WiFi-powered tablets that displayed GPS maps with their locations. In a future real-life scenario, the medics would study this information in an ambulance racing to the scene, getting a head start on triaging victims and determining whom to treat first.
“If a medic can get to certain patients in 10 minutes they’ll live; if not, they’ll die,” said Derek A. Paley, the Willis H. Young Jr. Professor of Aerospace Engineering Education and leader of the UMD team, named RoboScout, comprising more than two dozen students, researchers and faculty.
The simulation was a preview of what’s to come starting Friday, when RoboScout will compete as a finalist in the 2023-26 Defense Advanced Research Projects Agency (DARPA) Triage Challenge in College Station, Texas. Fourteen university and industry teams will be judged on how well their high-tech systems could assist military and civilian first responders, vying for a $1.5 million grand prize in the weeklong competition sponsored by the Pentagon’s advanced technology R&D agency.
For healthcare workers reacting to school shootings, military battles and highway pileups, AI assistance could be a boon, not just with triage but by providing real-time information to surgeons standing by at hospitals.
“This technology is important for homeland security,” said Jody Cantu, the Air Force Research Laboratory’s En Route Care technical lead that oversees the research programs for transport and care for airmen injured in battle. Her team is in contracts with Paley to develop a RoboScout-like system for situations where airmen might have been exposed to dangerous chemicals in a warzone.
“That would make it a mass casualty scenario, but you don’t want your medics to become casualties, too, so we’d deploy the RoboScout platform first,” she said.
Paramedic James Martin attends to a role-player posing as a victim during a simulation at the Fairfax Urban Search and Rescue’s training facility in Lorton, Va., in July.
Fifteen minutes before he was plugging Tammy’s wound, Martin had been on the sideline studying his tablet as the drones and robotic dogs searched for victims. With AI-equipped sensors that include color and thermal cameras, radar and microphones, the dogs hear screams, converse with victims and determine vitals like heart rate, alertness and respiratory distress along with injuries like blood loss, trauma and amputation.
One dog robot, equipped with cameras, two-way audio and a large-language model that allows it to hold a conversation, approached a bloodied victim and said, “I’m here to help. Move your arms so I can check if you’re mobile.” After taking pictures, the dog said: “I need to assess others. A medic will be here soon.”
As that happened, color-coded pushpins appeared on Martin’s map: red for high-need, yellow for lower-need, black for dead.
“I see the first patient. It’s yellow, no images yet,” Martin told Vienna from the sideline. “Now I have an urgent — someone’s hemorrhaging, respiratory distress.” A red pin popped up. “Amputation. Oculars closed. Not speaking or moving.”
Each victim was assigned a number. “You go to 22, I’ll go to 1,” Martin told Vienna moments before they finally sprinted onto the course.
In his 19-year career, Martin has seen victims of gunshot wounds, cardiac arrests and diabetic attacks die before his eyes; some would have lived if he’d arrived earlier. “I might be doing CPR or starting an IV, and they code,” said Martin, who works for the trauma resuscitation unit of the University of Maryland Medical Center’s R Adams Cowley Shock Trauma Center in Baltimore.
During the DARPA competition, RoboScout will square off against finalists representing schools like the University of Pennsylvania and Carnegie Mellon University, as well as national and international private companies like Battelle. RoboScout is one of five DARPA-funded teams. Representatives of venture capital firms exploring commercialization possibilities for autonomous technologies are expected to attend.
Zachary Bortoff, an aerospace engineering doctoral student from Avon, Conn., who helped write AI code for the three dogs’ sensors, is pleased by their performance so far.
“It’s cool to watch your autonomous robots do what they’re programed to do, to identify and approach a person and scan them,” he said. “Encoding that into a computer is very abstract. You can’t predict how a robot will respond to weird environments.”
The mock highway pileup was staged by the C-STARS Baltimore Sim Lab at the Shock Trauma Center, which trains Air Force medics for deployment. Air Force Research Laboratory volunteers served as role players.
Visitors included representatives of the Air Force, Maryland Civil Air Patrol, Maryland National Guard and UMD Maryland Fire and Rescue Institute, who watched Martin and Vienna race around the football field-size course, keeping in communication — “You go over to that yellow! I got this red!” Martin yelled as he approached a man slumped over the wheel of a moving truck with an apparent head injury. As they hopscotched the course, the medics eventually started treating patients with less-severe injuries.
After 10 minutes, a team member yelled, “Terminate!” As he trudged off the grounds, Martin praised the technology, noting that Tammy, the second victim he treated, would not have survived her groin injury without immediate care.
“Being able to do an assessment from afar, with live pictures, was game-changing,” he said.
A RoboScout team member observes drone footage of a highway pileup simulation. Two drones and three robot drones used A.I. to identify victim location, vial signs and type and severity of injuries.
UMD Research Changes Lives
At the University of Maryland, scientists and scholars come together to spark new ideas, pursue important discoveries and tackle humanity's grand challenges—improving lives in our communities and across the globe. See more examples of how UMD research changes lives at today.umd.edu/topic/research-impact.